AI gross margin, explained for founders
What investors mean by gross margin for an AI product, how inference costs move it, and how to report it with a straight face.
By Keyveve, Inc. · Updated 2026-10-22
For classic SaaS, gross margins of 75–85% were taken for granted. AI products often land lower because every request has a real marginal cost. Investors now ask about gross margin earlier and in more detail.
The formula
Gross margin = (revenue − cost of revenue) ÷ revenue.
For an AI product, cost of revenue typically includes the inference that serves customers in production, hosting and the infrastructure directly serving the product, third-party APIs used per request, and customer support. It doesn't include R&D inference: experiments, evaluations and staging.
Why classification matters
Two companies with the same bill can report very different margins depending on whether development inference sits in cost of revenue. Be consistent and be able to show your work:
- tag calls with a
feature(or aprojectfor internal work) and anenvironment:development,staging,testandresearchcount as R&D by default, everything else as cost of revenue; - classify by project, feature and environment, not by guesswork at month end;
- keep a written method.
Margin per customer
Company-level margin hides the shape that matters: a handful of heavy users on a flat plan can erase the margin of hundreds of light ones. Margin per customer (revenue from billing minus the inference that customer caused) shows which plans need limits, usage pricing or a cheaper model.
Questions you'll be asked
| Question | What to have ready |
|---|---|
| What's gross margin with and without credits? | Usage priced at list rates, credits shown separately |
| How does margin change with scale? | Cost per customer and per active user over time |
| What happens if model prices drop or rise? | Spend by model, and what a switch would cost |
| Are there unprofitable customers? | The list, and what you're doing about them |
With Veqen
Connect Stripe (Pro or Team) and your provider cost APIs; the Customers page shows revenue, AI cost and margin per customer (all AI cost tagged to the customer, including development calls, so tag those separately), and the monthly close gives the classified cost of revenue with its method.
Veqen records what every AI call cost and who it was for, reconciles it to your provider bills and produces the journal entries. Start free or read the quickstart.